17 Feb 2017 - Mehrnoosh Sadrzadeh: Seminar
Entailment in Compositional Distributional Semantics
Distributional models of meaning represent words by vectors, obtained from the frequency of their occurrences in corpora of text. Similarity is the main relation applied to word vectors, but one can also do entailment using inclusion relations over features of these vectors. This has indeed been studied before, e.g. see Kotlerman et al, Herbelot and Ganesalingam, and Rimell. In recent joint work with D. Kartsaklis, presented at COLING and LACL 2016, we showed how inclusion relations can be lifted compositionally from word features to phrase and sentence features. We considered and compared performances of different compositional models: from simple addition and multiplication, to min and max, to matrix multiplication and cube contraction. In an earlier version (joint with E. Balkir) we also worked with density matrices. We experimented on datasets containing verb and adjective noun phrases, and SVO sentences. I will present the setting and our results. I will then discuss how one can develop logic for distributional semantics based on this and a few other possible entailment relations. The logic part is on going work with J. Hedges, M. Dostal and R. Muskens.
I got a BSc in Computer Software Engineering and an MSc in Logic from Sharif University, Tehran, Iran. After obtaining a PhD in 2006 with joint supervision at UQAM and Oxford and two postdoctorates in Southampton and Paris; in 2008 I obtained an EPSRC PDRF and a JRF in Wolfson College in Oxford; in 2011 I received an EPSRC CAF in Oxford; in 2013, a I got a lectureship at Queen Mary University London, where at the moment I am a senior lecturer and teach NLP and mathematics for engineers. I have worked on algebra and proof theory for applied logical systems such as type-logical systems for natural language and on vector composition and algebras for distributional semantics. I have done recurrent PC work in relevant conferences and workshops, including serving in the committee of the Beth dissertation award and semantic co-chairship of ACL.